Research And Application Of Recommendation Algorithm Based On Convolutional Neural Network And Matrix Factorization

With the explosive growth of Internet information,information overload is becoming more and more serious,and it is more and more difficult for people to find the target information using traditional methods,so the recommendation system came into being.Especially the recommendation system represented by collaborative filtering algorithm has attracted much attention.Collaborative filtering algorithm generates personalized recommendation services for users with similar interests through user rating information based on the real world characteristics of "people clustering and things clustering ".However,due to the limitation of sparse rating data and cold start,collaborative filtering algorithm has low recommendation accuracy.Therefore,this thesis focuses on the bottleneck of collaborative filtering algorithm.(1)This thesis proposed a hybrid collaborative filtering recommendation algorithm combineing two collaborative filtering algorithms based on users and items.It mainly improves the calculation method of calculating the similarity between users or between items.By adding a common scoring penalty factor,it can alleviate the common scoring when calculating the similarity and reduces the error of similarity calculation.At the same time,the time adaptive weight function is introduced to solve the problem of user interest drift.By verifying the algorithm in the public movie data set Movie Lens,the experimental results shows that the MAE value is reduced to 0.71 and the accuracy of score prediction is improved by 5.7% compared with the collaborative filtering algorithm.(2)This thesis also put forward a personalized recommendation algorithm combining convolutional neural network and matrix decompositio.By using the convolutional neural network to learn the characteristics of users and items and extracting the hidden feature matrix information,the implicit feature matrix information is replaced in matrix decomposition,and then fitting predicted scores,thus greatly ease the recommendation effect caused by the data sparseness problem and improve the user experience.the recommendation effect caused by the data sparse problem is not good,but the user experience is improved.Simultaneously,the algorithm is simulated in the Movie Lens public data set.The results shows that the improved recommendation model proposed in this paper has a good recommendation effect,which RSME is reduced to 0.81.(3)Based on the above work,the thesis designs and develops a recommended system prototype,which provides users with a personalized recommendation platform for movies and has a good application value for effectively alleviating information overload and improving user experience.